Parameter optimisation for texture completion

In 3D reconstruction applications frequently large mesh sections are missing. While the geometry can be created with mesh completion techniques, reconstructing the texture is more difficult. An exemplar-based synthesis usually requires that the missing region contains only one texture pattern. Image...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Nguyen, Hoang M., Wunsche, Burkhard C., Delmas, Patrice, Lutteroth, Christof
Format: Tagungsbericht
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In 3D reconstruction applications frequently large mesh sections are missing. While the geometry can be created with mesh completion techniques, reconstructing the texture is more difficult. An exemplar-based synthesis usually requires that the missing region contains only one texture pattern. Image inpainting techniques work for more complex textures, but were designed primarily to fill narrow missing regions and tend to produce undesirable results when the missing region is large. We solve this problem by exploiting the fact that most 3D objects contain different surface regions with similar textures. We use an appearance space to identify suitable texture patches fitting with the boundary of the missing region, and fill the region by using a patch-based synthesis combined with Poisson interpolation. We evaluate the effect of different parameters and demonstrate its improved performance over existing inpainting techniques.
ISSN:2151-2191
2151-2205
DOI:10.1109/IVCNZ.2013.6727020